Lehre/Seminar Metaphor Processing in Natural Language (Master)/en
Advanced topics in Natural Language Processing - Metaphor Processing in Natural Language (Master)
| Type of course | seminar |
| Lecturer(s) | Harald Sack |
| Instructor(s) | Mehwish Alam |
| Subject | |
| Credit Points | 3 |
| Control of Success | |
| Term | summer |
You find additional information, the time schedule and room numbers in the University Course Overview.
Course Overview https://campus.studium.kit.edu/events/catalog.php#!campus/all/event.asp?gguid=0x9A32C40C85E041039FE1C65263791B46
Student Portal https://portal.wiwi.kit.edu/ys/4453/
A metaphor is a cognitive operation involving usage of natural language and crossdomain conceptual mapping. The idea behind conceptual mapping comes from Conceptual Metaphor Theory defined by Lakoff and Johnson in Metaphors We Live By in 1980 [1] which implies that humans create metaphors by mapping the properties of a concept from one domain to another (cross domain conceptual mapping), e.g., “sweet person” which says that the niceness of the person is mapped to the sweetness of the sugar since a person does not taste sweet but sugar does. Processing metaphors has become one of the fundamental challenges for machine understanding. More specifically, it can be difficult for conversational agents (such as Alexa, SIRI, etc.) to understand the intended meaning behind the uttered metaphorical expressions, e.g., “My car drinks petrol, what should I do?”. Another challenging aspect would be in the field of machine translation where metaphors would be translated literally instead of interpreting the correct meaning of the metaphor and then performing translation.
Metaphor identification and interpretation is not the only challenge in processing metaphors. The machines should also be able to generate metaphorical expressions for adding a kick to the generated answers, for example, in the case of conversational agents. It can also be used for machine translation, for example, interpreting the metaphor in one language and then generating the appropriate corresponding metaphor in another language. To date many algorithms have been proposed for metaphor identification and interpretation. These algorithms range from hand crafted rules to machine/deep learning approaches. Moreover, many studies have also considered taking into account knowledge represented in existing repositories such as MetaNet.
In this seminar, we focus on an in-depth study of different state of the art algorithms for metaphor identification and interpretation in text as well as images, textual metaphor generation, selectional preferences for the plausibility of semantics of text, etc.
Contributions of the students:
Each student will be assigned at max 2 papers on the topic, out of which the student will have to give a seminar presentation and write a seminar report paper of 15 pages explaining the methods from at least one of the two assigned papers, in their own words.
Implementation:
If the code is available from the authors, then re-implementation of the code for small scale experiments using Google Colab is required.
Temporary Timeline:
1st Meeting: Introduction and Course Organization - 14 April 2021
2nd Meeting: Paper Assignment - 21 April 2021
3rd Meeting: On student’s demand - 28 April 2021
4th Meeting: Student Presentations - 19 May 2021
5th Meeting: Student Presentations - 26 May 2021
6th Meeting: Student Presentations - 2 June 2021
7th Meeting: Student Presentations - 9 June 2021
8th Meeting: Student Presentations - 16 June 2021
Code submission: 25 July 2021
Seminar report submission: 25 July 2021
Two students will present on each of the Presentation Meetings.
[1] George Lakoff and Mark Johnson. Metaphors we Live by. University of Chicago Press, Chicago, 1980.
[2] Ellen Dodge, Jisup Hong, and Elise Stickles. MetaNet: Deep Semantic Automatic Metaphor Analysis. In Proceedings of the Third Workshop on Metaphor in NLP, pages 40–49. Association for Computational Linguistics, 2015.